Google is rolling out Genie 4 to eligible Google AI Ultra subscribers, bringing live sound and dialogue to the playable worlds they generate.
Inside Project Genie, you can now type a prompt while exploring to change the weather, add a character, or redirect the action without starting over. Google AI Pro members can join a waitlist on the same page, but they can’t use Genie 4 yet.
Diego Rivas, group product manager at Google DeepMind, and Pavan Kumar, group product manager at Google Labs, announced the rollout in a Google blog post. The post doesn’t name any countries, so it doesn’t establish whether Ultra subscribers in the UAE can try the new model.
What Genie 4 adds and why Genie 3 is still there
Genie 4 is the first in the series to generate audio alongside its visuals in real time. Ambient sound and character dialogue adapt as the scene unfolds. The Project Genie agent takes a description in everyday language and coordinates Google’s models to generate the starting visuals, while Gemini can suggest where to go or trigger events.
Genie 3 stays in the model picker for longer runs and for starting a world from a 360-degree panorama. It now lets you explore for up to three minutes per generation, up from 60 seconds, and link multiple worlds for longer journeys. Google DeepMind’s model page says Genie 4 supports a few minutes of continuous interaction, not extended sessions.
MarkTechPost reported that Genie 3 ran at 720p and 24 frames per second, with about one minute of visual memory. Google’s rollout post gives no equivalent figures for Genie 4.
A new community gallery lets users publish, explore, and remix worlds. Google’s selected examples include Villa Giallo, a room-by-room mystery set in a Gothic manor, made with Will aka BLVCKL!GHT. The Life of Ants was built with filmmaker Anand Gandhi. All video and audio generated in Project Genie carries a SynthID watermark, and Google screens all model inputs and outputs for safety.
What still breaks the simulation
Genie 4 is built on Gemini. In its 27 October research preview, Google DeepMind explained that Gemini Omni models real-world appearance and behaviour, while other Gemini models help keep scenes consistent.
Visual details can drift and objects don’t always stay consistent, especially as a session runs longer, after several complex edits, or as you move around. The model can’t recreate real-world locations with perfect accuracy.
Google still treats both Genie models as research. It lists camera lag, complex physical interactions, precise real-time audio, and keeping scenes consistent over longer runs among the problems it’s working on.
The October preview began with trusted testers and named early-access partners of the Gemini API and Google AI Studio alongside Ultra subscribers as the next groups. Google hasn’t published API pricing or a general availability date.


















